Transitions between biomes are common and directional in Bombacoideae (Malvaceae)

To quantify evolutionary transitions between tropical evergreen rain forest and seasonally dry biomes, to test whether biome transitions affect lineage diversification and to examine the robustness of these results to methodological choices.


| INTRODUC TI ON
Evolutionary transitions among biomes have been suggested as critical for the generation of plant diversity (Donoghue & Edwards, 2014). In particular, a high connectivity of biota among biomes in tropical America (the Neotropics) might have been essential for the assembly of its globally outstanding diversity (Antonelli et al., 2018;Zizka, 2019). Biomes are vegetation units defined by functionally similar plant groups and similar environmental conditions (Moncrieff, Hickler, & Higgins, 2015). Although the definition, meaning and delimitation of biomes remain controversial (Moncrieff et al., 2015;Mucina, 2019), they are often used to understand how broad-scale ecological niches change in evolutionary time (e.g. Bacon, 2013).
Generally, species tend to retain their ancestral ecological niche over time (Wiens & Donoghue, 2004). Therefore biome transitions -the shift of evolutionary lineages into new biomes -have been considered rare, especially in some biomes such as seasonally dry tropical forests (Crisp & Cook, 2012;Gagnon, Ringelberg, Bruneau, Lewis, & Hughes, 2019). Environmental dissimilarity might be a major factor limiting interchange of plant lineages among biomes (Crisp et al., 2009) and within the frost-free tropics, seasonal water availability may be particularly important constraining factor (Hughes, Pennington, & Antonelli, 2013;Olmstead, 2013).
Three major lowland tropical biomes are evergreen tropical rain forests, seasonally dry tropical forest and savanna (including tropical grass-and shrublands). Despite key differences in some ecological drivers (e.g. fire frequency), savanna and seasonally dry forest share a marked seasonality in water availability, which sets them apart from evergreen rain forest (Pennington, Lehmann, & Rowland, 2018). This raises a broader question of transitions into and out of regions with a seasonally dry climate .
Evolutionary adaptations to any new biome may be difficult to evolve, but once a lineage is established, one might expect it to diversify (Moore & Donoghue, 2007), for instance, due to the release from competitors and herbivores, or adaptive radiation in the new environment. Indeed, a link between biome transitions and increased diversification has been found in some taxa such as Fabaceae (Koenen et al., 2013), Proteaceae (Onstein et al., 2016) and Malvaceae (Areces-Berazain & Ackerman, 2017), but not in others, for example, Protea (Valente et al., 2010). In the Neotropics, the repeated adaptation to seasonally dry habitats has been hypothesized as an important mechanism of plant diversification, in particular, in savannas (Areces-Berazain & Ackerman, 2017), and in the Paleotropics aridification and transitions to seasonally dry biomes seem to have increased diversification in some groups (Abrams et al., 2019;García-Aloy et al., 2017). Little is known for seasonally dry forests, (Dexter et al., 2015), which seem to be relatively isolated through time and space (Pennington & Hughes, 2014).
In summary, while biome transitions are an integral process in the assembly of tropical biodiversity, there is much we do not know about transitions to seasonally dry biomes and their impact on lineage diversification. Here, we quantify the number of biome transitions in the Bombacoideae, a subfamily of the Malvaceae with a centre of diversity in the Neotropics. We test two hypotheses related to the effect of biome transitions among evergreen rain forest and seasonally dry biomes on diversification.
1. Transitions from evergreen rain forests into seasonally dry biomes occur multiple times and are more common than viceversa. We expect this because of the older age of evergreen rain forests and results from prior studies identifying savannas as lineage sinks, especially in the Neotropics (Donoghue & Edwards, 2014;Freitas, Bacon, Souza-Neto, & Collevatti, 2019;Pennington & Hughes, 2014).
2. Transitions from evergreen rain forest to seasonally dry biomes increased net diversification due to competitive release and adaptive radiation. An increased diversification after biome transitions has been documented in other lineages (Cardillo et al., 2017;Souza-Neto, Cianciaruso, & Collevatti, 2016).

| Study group
The Bombacoideae is distributed across all tropical regions, with approximately 90% of its species richness in the Neotropics, thus reflecting the general pattern of globally outstanding plant diversity in this region. Furthermore, the Bombacoideae occur in a variety of different habitats suggesting multiple biome transitions ensuring a sufficiently large sample to estimate directionality in transitions.
We followed the taxonomy of Bombacoideae by Robyns (1963) updated with recent revisions when necessary (www.tropi cos. org). In total, our list comprised 174 accepted species in 17 genera (Appendix S1).

| Phylogenetic reconstruction and dating
We obtained DNA sequences from two nuclear (ETS and ITS) and three chloroplast markers (matK, trnS-trnG and trnL-trnF) for 103 species (59% of Bombacoideae, including all genera) from Carvalho-Sobrinho et al. (2016). We aligned the sequences using MUSCLE v3.6 (Edgar, 2004) with manual adjustment (Simmons, 2004). We partitioned the nuclear data by locus to allow for variation in substitution models and analysed the chloroplast markers as a single unit separate from the nuclear genes. The use of fossils allows for a minimum constraint on a clade's age, where non-uniform (e.g. exponential) prior probability densities are most often used (e.g. Warnock, Parham, Joyce, Lyson, & Donoghue, 2015). We dated the phylogeny using two fossil calibration points: (a) a macro-fossil of Malvaciphyllum macondicus (Wing, Herrera, Jaramillo, Gómez-Navarro, & Labandeira, 2009) Figure   S2.1 in Appendix S2).

| Geographical distribution
We compiled geographical localities of Bombacoideae from our own fieldwork and public databases (www.gbif.org, biendata.org, Gilles et al., 2016 et al., 2017). We used the rgbif package v1.1.0 (Chamberlain, 2017) in R (R Core Team, 2019) to obtain records from www.gbif.org (GBIF. org, 2018). We only included records filed as the accepted species names and used the taxize R-package v0.9.5 (Chamberlain & Szöcs, 2013) to resolve spelling errors in the species names. We merged sub-specific ranks under the accepted species name, and restricted species' occurrences to the native species range on a regional level based on our field experience and the literature. We retained only one record per species per site and cleaned occurrence records geographically using the CoordinateCleaner R-package v2.0-7 (Zizka et al., 2019). To visualize the global species richness of Bombacoideae, we generated species ranges from the occurrence records using geospheric convex hulls clipped to coastlines using the CalcRange function of the speciesgeocodeR R-package v2.0-10 (Töpel et al., 2017), using a 50km buffer for species with less than 3 occurrences.

| Biome classification
Based on the occurrence records and a widely used global biome definition (Olson et al., 2001), we classified species as: evergreen rain forest present in 'Tropical and subtropical moist broadleaf forests'; seasonally dry forest present in `Tropical and subtropical dry broadleaf forests' or 'Deserts and xeric shrublands'; or savanna present in 'Tropical and subtropical grasslands, savanna and shrubland'.
To account for outlier individuals and imprecision in geographical coordinates, we counted a species as present in a biome, if at least 5% of its records occurred there. We treated seasonally dry forest and savanna differently for the ancestral state reconstruction, because they differ in their ecology (Pennington et al., 2018, i.e. the presence of fire) and because they might differ in their affinities with evergreen rain forest. However, since we were interested in transitions among evergreen rain forest and seasonally dry biomes we combined these two biomes as seasonally dry biomes (SDB) for the estimation of diversification rates. We justify this with the potential importance of rainfall seasonality in the diversification of flowering plants (Areces-Berazain & Ackerman, 2017) and because Bombacoideae are often used as indicator species of evergreen rain forest in the fossil record (Morley, 2000). Furthermore, a more fine-scale biome classification would lead to reduced statistical power and classification accuracy of species to biomes (Silva de Miranda et al., 2018).

| Ancestral biome estimation
We used biogeographical stochastic mapping based on the dispersal-cladogenesis-extinction model (DEC) as implemented in BioGeoBEARS v1.1.2 (Dupin et al., 2017;Matzke, 2016) to reconstruct ancestral biomes on the phylogeny. Because there is good evidence for an older age of evergreen forests as compared to savanna and seasonally dry tropical forest, as well as fossil evidence that the Bombacoideae are ancestrally a rain forest group (Wing et al., 2009), we used a time-stratified model together with an areas-allowed-matrix, and limited the group to evergreen rain forest before the Miocene (23.03 mya). We used 1,000 stochastic replicates on the maximum clade credibility tree from the BEAST analysis. To quantify the number of biome transitions (Hypothesis 1), we counted the number of transitions from evergreen rain forest into either seasonally dry biomes and vice versa, inferred by the biogeographical stochastic mapping.

| Diversification rate estimation
We used the GeoHiSSE model (Caetano, O'Meara, & Beaulieu, 2018) as implemented in the GeoHisse function of the hisse v.1.9.6 R-package (Beaulieu & O'Meara, 2016), to estimate state-specific diversification and extinction rates from the phylogeny, and hence the impact of biomes on diversification (Hypothesis 2). GeoHiSSE estimates speciation and extinction rates dependent on geographical trait states, as well as transition rates among states while allowing for widespread ancestors and taking sampling frequencies into account. We chose GeoHiSSE above other SSE methods, since it (a) can account for widespread species, (b) can include 'concealed traits' and therefore is less prone to false positives (Caetano et al., 2018), and (c) has a limited number of parameters suited for our moderate-

| Reliability of the results
We tested the sensitivity of our conclusions to four potential caveats and sources of uncertainty. See Appendix S4 for more detail on the tests related to the ancestral biome reconstruction and Appendix S5 for more detail on the tests related to diversification rate estimation.

| Temporal evolution and recent biogeography
We inferred the root age of the Bombacoideae between 53.5 and 59.3 mya (Ma), close to the fossil constraint for the crown node of Bombacoideae + Malvoideae (58 Ma). Most branches in the reconstructed phylogeny were well supported, with some exceptions in Ceiba, Eriotheca and Pachira ( Figure S2.3 in Appendix S2).
We found 14,865 high-quality occurrence records for 172 species (98% of the Bombacoideae; max. 3,062 records for C. pentandra; median of 19 per species; Figure S2.4, Appendix S2 for geographical sampling intensity; Appendix S3 for species range maps). The range maps confirmed Amazonia and the Atlantic forest as centres of Bombacoideae diversity (Figure 1).

| Diversification rate analyses
We found no significant effect of biome state on the diversification rates. The best-fitting GeoHiSSE model had no trait-dependent speciation, unequal transition rates and one concealed trait. The lack of significant difference in diversification rates between EFB and SDB in the Bombacoideae was robust to phylogenetic uncertainty and biome classification (Table 2) (Table 2). While our main analyses, and the additional HiSSE analyses suggested higher transition rates from SDB to EFB, the direction was reversed when using a phenology-based biome definition. For phylogenetic uncertainty and biome classification, the direction was variable and the absolute diversification and transition rates varied by orders of magnitude among the individual replicates ( Figure   S5.10 in Appendix S5).

| D ISCUSS I ON
Here, we tested hypotheses on the role of biome transitions in the evolutionary of the tropical plant group Bombacoideae. We found support for multiple independent transitions among the evergreen rain forest (EFB) and seasonally dry biomes (SDB) and more transitions from EFB to SDB than vice-versa (Hypothesis 1). We rejected the hypothesis that biomes differ in net diversification (Hypothesis 2).

| Temporal evolution
The root-age of Bombacoideae is consistent with the dating of its sis- lineages now considered Malvoideae). Furthermore, our use of exponential priors on the two macrofossil calibrations constrained by the age of the strata they were derived from implies high confidence in their age, which seems justified in this case because the fossil specimens are well-preserved and present clear morphological synapomorphies allowing for their assignment to a specific nodes on the phylogeny. Additionally, the stratum the Malvaciphyllum fossil is derived from is temporally well-defined (Wing et al., 2009). Using less informative priors, at least with respect to maximum age, would likely shift divergences in the group to somewhat older ages.
The Bombacoideae have a relatively rich fossil record compared to other plant groups, especially for pollen (Jaramillo, Rueda, & Torres, 2011). Unfortunately, it is generally difficult to place pollen fossils in relation to recent taxa, due to high homoplasy in Malvaceae pollen morphology (Saba, 2007), making the full integration of fossil and molecular data challenging. However, our results on the number and direction of biome shifts should be robust to uncertainties in phylogenetic dating.

| Biome transitions
Our findings suggest caution using Bombacoideae fossils as indicators of past evergreen rain forests (e.g. Morley, 2000;Pross et al., 2012), unless these fossils have biome-specific traits. The number TA B L E 1 Biome transitions in Bombacoideae among evergreen rain forest (EFB) and seasonally dry biomes (SDB), the latter including seasonally dry forest (SDF) and savanna (SAV)

F I G U R E 2
Biome evolution in Bombacoideae. The evergreen rain forest is represented in purple, seasonally dry forest ('Seasonally dry tropical and subtropical forest' and 'Deserts and Xeric Shrublands', Olson et al., 2001) in blue and savanna in yellow. Dispersals into seasonally dry forest and savanna were allowed from the beginning of the Miocene onward based on fossil information. There are multiple independent transitions into seasonally dry biomes, especially in the last 10 million years of biome transitions among EFB and SDB are high, at least at the large scale ( Figure 2, Table 1), especially compared to other similarsized groups (Cardillo et al., 2017;Estrella et al., 2017), which is likely partly due to how we chose to account for widespread species and to our use of biogeographical stochastic mapping, which reconstructs anagenetic events along branches, rather than just counting shifts observed at nodes (which represent the minimum number of shifts).
Our results from the biogeographical stochastic mapping suggest more transitions towards seasonally dry biomes than the reverse. This fits with expectations based on the age of the biomes and observations from other lineages (Pennington & Hughes, 2014;Simon & Pennington, 2012). The evergreen rain forests of Amazonia -one of the diversity centres of Bombacoideae -have been suggested as a regional and global species source, 'pumping lineages' into other biomes (Antonelli et al., 2018(Antonelli et al., , 2015 Model choice partly affected the ancestral biome reconstruction, with GeoHiSSE suggesting the majority of lineages as generalist rather than rain forest specialist ( Figure S4.6 in Appendix S4).
This seems unlikely and might be due to the unconstrained treat- and specimen data are more reproducible, less influenced by researcher biases, and better suited to computational investigations of uncertainty in downstream analyses, than expert-based TA B L E 2 Evolutionary transition and diversification rates among evergreen rain forest (EFB) and seasonally dry biomes (SDB), the latter including seasonally dry tropical forest (SDF) and savanna (SAV) in Bombacoideae

| Biomes and diversification
We did not find a significant relation between diversification rate and biome type, rejecting the hypothesis that biome transitions into seasonally dry biomes are a driver of increased diversification, although this finding is subject to uncertainty. This could reflect a lack of statistical power or indicate that other factors besides the adapta- The transition and diversification rate analyses were sensitive to all types of tested uncertainty. This was partly expected due to the different specifications of the models (especially if they account for generalist species) and the resulting changes in the classification of tip trait states, as well as the relatively small size of our phylogeny (Davis et al., 2013). However, these results were also surprising, especially concerning sensitivity to model choice on the qualitative conclusions (Table S5.2, Appendix S5) as well as the effect of phylogenetic uncertainty and biome classification on the estimated rates ( Figure S5.10, Appendix S5). Reasons for the sensitivity of the diversification rate analyses to model choice and uncertainty deserve more study, but the results indicate that we currently lack evidence that biome alters diversification rate in Bombacoideae.

| CON CLUS IONS
We show that evergreen rain forest -seasonally dry biome transitions are common in Bombacoideae, especially within the Neotropics. These findings are robust to methodological choices and support the view that the evergreen rain forest-seasonally dry biome boundary is permeable for this plant lineage on evolutionary time scales. Furthermore, this permeability is directional with transitions from evergreen rain forest to seasonally dry biomes being more common than the reverse.
Our results also demonstrate that model choice as well as different biome classifications and biome definitions can lead to qualitatively different conclusions, stressing the importance of carefully selecting a biome-scoring scheme that is suitable for the question at hand and testing its sensitive to methodological choices and assumptions.

DATA ACCE SS I B I LIT Y S TATE M E NT
A species list is available in the Supporting material (Appendix S1). The DNA alignments, the shapefiles of species ranges and all analyses scripts with input data are available from zenodo (https:// zenodo.org/recor d/26343 08#.Xku30 0p7mHt)

ACK N OWLED G EM ENTS
We thank two anonymous reviewers and the subject editor for helpful comments on the manuscript. We thank Francisco Velasquez for preliminary analyses and discussion of the results, Søren Faurby for help with analyses in an earlier version of the manuscript and Ivana Kirchmair for helpful comments on the manuscript. We thank C. Chatelain for sharing occurrence data from the Africa plant database, Jan Wieringa for sharing records from WAG and all data contributors to GBIF and species link for their effort in collecting, digitizing, and providing specimens used in this study. AZ is thankful for funding by iDiv via the German Research